, 18:30–20:00
AI Safety Weekly: Alignment from Token Zero
V. Moskvoretskii (EPFL) on building an assistant persona into pretraining rather than bolting alignment on afterwards — a technique now part of training Apertus 2, Switzerland’s open model. Moderate technical background assumed.
- Dates
- Time
- 18:30–20:00
- City
- Zurich
- Venue
- ETH Main Building, HG D3.3, Rämistrasse 101
- Organiser
- Zurich AI Safety
- Format
- Meetup
- Cost
- Free (registration by approval, food provided)
- Topics
- AI safety, Open-source LLMs, AI research
- Verified via
- luma.com